diff --git a/Core/Config.cpp b/Core/Config.cpp
index 520002dd73..4fee46c4b7 100644
--- a/Core/Config.cpp
+++ b/Core/Config.cpp
@@ -172,6 +172,9 @@ void Config::Load(const char *iniFileName)
 	debugConfig->Get("FontWidth", &iFontWidth, 8);
 	debugConfig->Get("FontHeight", &iFontHeight, 12);
 
+	IniFile::Section *gleshacks = iniFile.GetOrCreateSection("GLESHacks");
+	gleshacks->Get("PrescaleUV", &bPrescaleUV, false);
+
 	KeyMap::LoadFromIni(iniFile);
 
 	CleanRecent();
diff --git a/Core/Config.h b/Core/Config.h
index 614c42fba0..aca6d83603 100644
--- a/Core/Config.h
+++ b/Core/Config.h
@@ -113,6 +113,19 @@ public:
 	int iTouchButtonOpacity;
 	float fButtonScale;
 
+
+	// GLES backend-specific hacks. Not saved to the ini file, do not add checkboxes. Will be made into
+	// proper options when good enough.
+	// PrescaleUV:
+	//   * Applies UV scale/offset when decoding verts. Get rid of some work in the vertex shader,
+	//     saves a uniform upload and is a prerequisite for future optimized hybrid 
+	//     (SW skinning, HW transform) skinning.
+	//   * Still has major problems so off by default - need to store tex scale/offset per DeferredDrawCall, 
+	//     which currently isn't done so if texscale/offset isn't static (like in Tekken 6) things go wrong.
+	bool bPrescaleUV;
+
+	// End GLES hacks.
+
 	// SystemParam
 	std::string sNickName;
 	int ilanguage;
diff --git a/Core/MIPS/ARM/ArmCompFPU.cpp b/Core/MIPS/ARM/ArmCompFPU.cpp
index 231a0f61b7..9af3490644 100644
--- a/Core/MIPS/ARM/ArmCompFPU.cpp
+++ b/Core/MIPS/ARM/ArmCompFPU.cpp
@@ -55,11 +55,11 @@ void Jit::Comp_FPU3op(u32 op)
 	case 1: VSUB(fpr.R(fd), fpr.R(fs), fpr.R(ft)); break; //F(fd) = F(fs) - F(ft); //sub
 	case 2: { //F(fd) = F(fs) * F(ft); //mul
 		u32 nextOp = Memory::Read_Instruction(js.compilerPC + 4);
-		// Optimise possible if destination is the same
-		if (fd == ((nextOp>>6) & 0x1F)) {
+		// Optimization possible if destination is the same
+		if (fd == (int)((nextOp>>6) & 0x1F)) {
 			// VMUL + VNEG -> VNMUL
 			if (!strcmp(MIPSGetName(nextOp), "neg.s")) {
-				if (fd == ((nextOp>>11) & 0x1F)) {
+				if (fd == (int)((nextOp>>11) & 0x1F)) {
 					VNMUL(fpr.R(fd), fpr.R(fs), fpr.R(ft));
 					EatInstruction(nextOp);
 				}
diff --git a/GPU/GLES/VertexDecoder.cpp b/GPU/GLES/VertexDecoder.cpp
index f149d06abf..a06cbb22ac 100644
--- a/GPU/GLES/VertexDecoder.cpp
+++ b/GPU/GLES/VertexDecoder.cpp
@@ -17,8 +17,9 @@
 
 #include "math/lin/matrix4x4.h"
 
-#include "../../Core/MemMap.h"
-#include "../ge_constants.h"
+#include "Core/Config.h"
+#include "Core/MemMap.h"
+#include "GPU/ge_constants.h"
 
 #include "VertexDecoder.h"
 #include "VertexShaderGenerator.h"
@@ -207,6 +208,27 @@ void VertexDecoder::Step_TcFloatThrough() const
 	uv[1] = uvdata[1];
 }
 
+void VertexDecoder::Step_TcU8Prescale() const {
+	float *uv = (float *)(decoded_ + decFmt.uvoff);
+	const u8 *uvdata = (const u8 *)(ptr_ + tcoff);
+	uv[0] = (float)uvdata[0] * (1.f / 128.f) * gstate_c.uScale + gstate_c.uOff;
+	uv[1] = (float)uvdata[1] * (1.f / 128.f) * gstate_c.vScale + gstate_c.vOff;
+}
+
+void VertexDecoder::Step_TcU16Prescale() const {
+	float *uv = (float *)(decoded_ + decFmt.uvoff);
+	const u16 *uvdata = (const u16 *)(ptr_ + tcoff);
+	uv[0] = (float)uvdata[0] * (1.f / 32768.f) * gstate_c.uScale + gstate_c.uOff;
+	uv[1] = (float)uvdata[1] * (1.f / 32768.f) * gstate_c.vScale + gstate_c.vOff;
+}
+
+void VertexDecoder::Step_TcFloatPrescale() const {
+	float *uv = (float *)(decoded_ + decFmt.uvoff);
+	const float *uvdata = (const float*)(ptr_ + tcoff);
+	uv[0] = uvdata[0] * gstate_c.uScale + gstate_c.uOff;
+	uv[1] = uvdata[1] * gstate_c.vScale + gstate_c.vOff;
+}
+
 void VertexDecoder::Step_Color565() const
 {
 	u8 *c = decoded_ + decFmt.c0off;
@@ -496,6 +518,13 @@ static const StepFunction tcstep[4] = {
 	&VertexDecoder::Step_TcFloat,
 };
 
+static const StepFunction tcstep_prescale[4] = {
+	0,
+	&VertexDecoder::Step_TcU8Prescale,
+	&VertexDecoder::Step_TcU16Prescale,
+	&VertexDecoder::Step_TcFloatPrescale,
+};
+
 static const StepFunction tcstep_through[4] = {
 	0,
 	&VertexDecoder::Step_TcU8,
@@ -638,21 +667,26 @@ void VertexDecoder::SetVertexType(u32 fmt) {
 		if (tcalign[tc] > biggest)
 			biggest = tcalign[tc];
 
-		if(g_DoubleTextureCoordinates)
-			steps_[numSteps_++] = throughmode ? tcstep_through_Remaster[tc] : tcstep_Remaster[tc];
-		else
-			steps_[numSteps_++] = throughmode ? tcstep_through[tc] : tcstep[tc];
-
-		switch (tc) {
-		case GE_VTYPE_TC_8BIT >> GE_VTYPE_TC_SHIFT:
-			decFmt.uvfmt = throughmode ? DEC_U8A_2 : DEC_U8_2;
-			break;
-		case GE_VTYPE_TC_16BIT >> GE_VTYPE_TC_SHIFT:
-			decFmt.uvfmt = throughmode ? DEC_U16A_2 : DEC_U16_2;
-			break;
-		case GE_VTYPE_TC_FLOAT >> GE_VTYPE_TC_SHIFT:
+		if (g_Config.bPrescaleUV && !throughmode && gstate.getTextureFunction() == 0) {
+			steps_[numSteps_++] = tcstep_prescale[tc];
 			decFmt.uvfmt = DEC_FLOAT_2;
-			break;
+		} else {
+			if (g_DoubleTextureCoordinates)
+				steps_[numSteps_++] = throughmode ? tcstep_through_Remaster[tc] : tcstep_Remaster[tc];
+			else
+				steps_[numSteps_++] = throughmode ? tcstep_through[tc] : tcstep[tc];
+
+			switch (tc) {
+			case GE_VTYPE_TC_8BIT >> GE_VTYPE_TC_SHIFT:
+				decFmt.uvfmt = throughmode ? DEC_U8A_2 : DEC_U8_2;
+				break;
+			case GE_VTYPE_TC_16BIT >> GE_VTYPE_TC_SHIFT:
+				decFmt.uvfmt = throughmode ? DEC_U16A_2 : DEC_U16_2;
+				break;
+			case GE_VTYPE_TC_FLOAT >> GE_VTYPE_TC_SHIFT:
+				decFmt.uvfmt = DEC_FLOAT_2;
+				break;
+			}
 		}
 
 		decFmt.uvoff = decOff;
diff --git a/GPU/GLES/VertexDecoder.h b/GPU/GLES/VertexDecoder.h
index 53b441e5e1..a2fa9ccce6 100644
--- a/GPU/GLES/VertexDecoder.h
+++ b/GPU/GLES/VertexDecoder.h
@@ -83,24 +83,19 @@ enum {
 };
 
 // Right now
-//   - only contains computed information
-//   - does decoding in nasty branchfilled loops
+//   - compiles into list of called functions
 // Future TODO
-//   - should be cached, not recreated every time
-//   - will compile into list of called functions
 //   - will compile into lighting fast specialized x86 and ARM
-//   - will not bother translating components that can be read directly
-//     by OpenGL ES. Will still have to translate 565 colors and things
-//     like that. DecodedVertex will not be a fixed struct. Will have to
-//     do morphing here.
 class VertexDecoder
 {
 public:
 	VertexDecoder() : coloff(0), nrmoff(0), posoff(0) {}
 	~VertexDecoder() {}
 
+	// prim is needed knowledge for a performance hack (PrescaleUV)
 	void SetVertexType(u32 vtype);
 	u32 VertexType() const { return fmt_; }
+
 	const DecVtxFormat &GetDecVtxFmt() { return decFmt; }
 
 	void DecodeVerts(u8 *decoded, const void *verts, int indexLowerBound, int indexUpperBound) const;
@@ -117,8 +112,13 @@ public:
 
 	void Step_TcU8() const;
 	void Step_TcU16() const;
-	void Step_TcU16Double() const;
 	void Step_TcFloat() const;
+
+	void Step_TcU8Prescale() const;
+	void Step_TcU16Prescale() const;
+	void Step_TcFloatPrescale() const;
+
+	void Step_TcU16Double() const;
 	void Step_TcU16Through() const;
 	void Step_TcU16ThroughDouble() const;
 	void Step_TcFloatThrough() const;
diff --git a/GPU/GLES/VertexShaderGenerator.cpp b/GPU/GLES/VertexShaderGenerator.cpp
index ee6f556c32..9acfe632dc 100644
--- a/GPU/GLES/VertexShaderGenerator.cpp
+++ b/GPU/GLES/VertexShaderGenerator.cpp
@@ -514,9 +514,15 @@ void GenerateVertexShader(int prim, char *buffer, bool useHWTransform) {
 
 		// Step 3: UV generation
 		if (doTexture) {
+			bool prescale = g_Config.bPrescaleUV && !throughmode && gstate.getTextureFunction() == 0;
+
 			switch (gstate.getUVGenMode()) {
 			case 0:  // Scale-offset. Easy.
-				WRITE(p, "  v_texcoord = a_texcoord * u_uvscaleoffset.xy + u_uvscaleoffset.zw;\n");
+				if (prescale) {
+					WRITE(p, "  v_texcoord = a_texcoord;\n");
+				} else {
+					WRITE(p, "  v_texcoord = a_texcoord * u_uvscaleoffset.xy + u_uvscaleoffset.zw;\n");
+				}
 				break;
 
 			case 1:  // Projection mapping.